1. Field of the Invention
The present invention relates to a hierarchical memory control system in a data processing system, and more particularly, to a hierarchical memory control system having a common global buffer storage unit provided between one or more central processing units each having a store-in buffer storage unit and one or more main storage units commonly used by the central processing units, in which the global buffer storage unit has a relatively large memory capacity in comparison with the memory capacity of each buffer storage unit in each of the central processing units.
2. Description of the Related Art
In a large-scale computer system, it is difficult to obtain satisfactory system performance with a single processor, and therefore, a multi-processor configuration is employed, to thus improve the system performance. In a data processing system having the multi-processor configuration, each processor has a buffer storage unit called a cache memory.
A buffer storage control system contained in a data processing system having a multi-processor configuration, a store-through system and a store-in system are known. In the store-through system, when data in a cache memory is modified by the central processing unit, corresponding data in the main storage unit is simultaneously modified. Conversely, in the store-in system, the data in the main storage is not modified at the time of the modification of the cache memory in the central processing unit, but the data modified by the central processing unit is copied into the main storage unit when the central processing unit requires the data in the cache memory to be transferred to the main storage or when the cache memory is full of data.
In the conventional data processing system, the store-through method is employed, but due to the increase of the number of the processors used in a multi-processor system, the storage accessing frequency is increased in the store-through method, because this method requires modified data storing steps. Therefore, the store-in method is now increasingly used in recent data processing systems.
On the other hand, to improve the system performance, it is essential to improve the storage accessing time and the storage throughput, and as one countermeasure, there is provided a configuration in which a medium speed, large capacity global buffer storage unit is provided between the cache memories in the central processing units and the main storage units.
In the conventional data processing system, however, when the global buffer storage unit is employed, only the store-through buffer storage control system is employed, as a store-in buffer control system can not be employed. Therefore, a problem arises in the conventional data processing system in that a sufficient improvement of the system performance is not obtained. Namely, in the conventional data processing system, when the global buffer storage unit is employed, the data in the global buffer storage unit is read by a central processing unit comprising a store-through cache memory, and thus the conventional data processing system does not support the store-in control system employed in most multi-processing systems.
Further, in the conventional data processing system comprising a global buffer storage unit, the size of each entry of the global buffer storage is the same as the size of each entry of the cache memory in the central processing unit. Therefore, in the conventional store-through system, a system such as that in which the size of the global buffer storage unit is larger than that of the cache memory does not exist. Further, in the conventional data processing system, a system such as that in which the storage capacity of the global buffer storage unit is larger than that of the cache memory does not exist.
Still further, in a conventional data processing system of a Tightly Coupled multi-processor (TCMP) type, which processes data by a common use of storage, cache memories in the central processing units can be mutually accessed. Nevertheless, a system such as that in which a global buffer storage unit is commonly used by the central processing units does not exist. Accordingly, in the conventional TCMP type data processing system, a problem arises in that a sufficient improvement of the system performance can not be obtained.